Eval network until the end for fairness

[blegat]

The others were evaluating the network completely, not just the gradient so it's fairer to do it in hand-cuda too.
Still the fastest though:

julia> HandCuda.neural(T, h, d, n; prealloc=false, gpu=false)
BenchmarkTools.Trial: 316 samples with 1 evaluation per sample.
 Range (min … max):   8.493 ms … 267.355 ms  ┊ GC (min … max):  0.00% … 96.10%
 Time  (median):      8.758 ms               ┊ GC (median):     0.00%
 Time  (mean ± σ):   15.838 ms ±  37.459 ms  ┊ GC (mean ± σ):  41.13% ± 18.44%

  █▂                                                            
  ██▄▁▁▁▄▆▅▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▆ ▅
  8.49 ms       Histogram: log(frequency) by time       261 ms <

 Memory estimate: 14.11 MiB, allocs estimate: 25.

julia> HandCuda.neural(T, h, d, n; prealloc=true, gpu=false)
BenchmarkTools.Trial: 424 samples with 1 evaluation per sample.
 Range (min … max):   4.810 ms … 284.633 ms  ┊ GC (min … max):  0.00% … 96.84%
 Time  (median):      8.260 ms               ┊ GC (median):     0.00%
 Time  (mean ± σ):   12.432 ms ±  32.054 ms  ┊ GC (mean ± σ):  35.17% ± 15.16%

  █▄                                                            
  ██▁▁▁▁▁▆▅▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▄▄ ▆
  4.81 ms       Histogram: log(frequency) by time       274 ms <

 Memory estimate: 8.55 MiB, allocs estimate: 19.

julia> HandCuda.neural(T, h, d, n; prealloc=false, gpu=true)
BenchmarkTools.Trial: 10000 samples with 1 evaluation per sample.
 Range (min … max):  215.534 μs … 286.947 ms  ┊ GC (min … max):  0.00% … 96.67%
 Time  (median):     225.652 μs               ┊ GC (median):     0.00%
 Time  (mean ± σ):   293.689 μs ±   2.987 ms  ┊ GC (mean ± σ):  14.14% ±  1.92%

       ▂█▅▁                                                      
  ▁▁▁▃▆████▇▅▄▄▃▂▂▃▃▂▂▂▂▂▂▁▂▂▂▂▂▂▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁ ▂
  216 μs           Histogram: frequency by time          291 μs <

 Memory estimate: 16.48 KiB, allocs estimate: 526.

julia> HandCuda.neural(T, h, d, n; prealloc=true, gpu=true)
BenchmarkTools.Trial: 10000 samples with 1 evaluation per sample.
 Range (min … max):  186.499 μs …  21.708 ms  ┊ GC (min … max): 0.00% … 28.60%
 Time  (median):     195.751 μs               ┊ GC (median):    0.00%
 Time  (mean ± σ):   222.872 μs ± 664.189 μs  ┊ GC (mean ± σ):  3.47% ±  1.16%

        ▂▄▆█▂                                                    
  ▁▂▂▃▄▇█████▆▅▅▃▃▃▃▃▂▂▂▂▂▂▂▃▃▂▃▃▂▂▂▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁ ▂
  186 μs           Histogram: frequency by time          244 μs <

 Memory estimate: 16.38 KiB, allocs estimate: 488.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 92.23%. Comparing base (7b8963c) to head (5504624).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main      #76   +/-   ##
=======================================
  Coverage   92.23%   92.23%           
=======================================
  Files          25       25           
  Lines        3219     3219           
=======================================
  Hits         2969     2969           
  Misses        250      250           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.